Droplet digital PCR-based detection of circulating tumor DNA from pediatric high grade and diffuse midline glioma patients

Abstract Background The use of liquid biopsy is of potential high importance for children with high grade (HGG) and diffuse midline gliomas (DMG), particularly where surgical procedures are limited, and invasive biopsy sampling not without risk. To date, however, the evidence that detection of cell-free DNA (cfDNA) or circulating tumor DNA (ctDNA) could provide useful information for these patients has been limited, or contradictory. Methods We optimized droplet digital PCR (ddPCR) assays for the detection of common somatic mutations observed in pediatric HGG/DMG, and applied them to liquid biopsies from plasma, serum, cerebrospinal fluid (CSF), and cystic fluid collected from 32 patients. Results Although detectable in all biomaterial types, ctDNA presented at significantly higher levels in CSF compared to plasma and/or serum. When applied to a cohort of 127 plasma specimens from 41 patients collected from 2011 to 2018 as part of a randomized clinical trial in pediatric non-brainstem HGG/DMG, ctDNA profiling by ddPCR was of limited use due to the small volumes (mean = 0.49 mL) available. In anecdotal cases where sufficient material was available, cfDNA concentration correlated with disease progression in two examples each of poor response in H3F3A_K27M-mutant DMG, and longer survival times in hemispheric BRAF_V600E-mutant cases. Conclusion Tumor-specific DNA alterations are more readily detected in CSF than plasma. Although we demonstrate the potential of the approach to assessing tumor burden, our results highlight the necessity for adequate sample collection and approach to improve detection if plasma samples are to be used.


Droplet digital PCR-based detection of circulating tumor DNA from pediatric high grade and diffuse midline glioma patients
The incorporation of tissue molecular profiling in patients with pediatric high-grade glioma (pHGG), diffuse intrinsic pontine glioma (DIPG) and other diffuse midline glioma (DMG) into clinical practice has been demonstrated to be essential to guide treatment decisions for these patients. 1,2 However, this requires invasive neurosurgical procedures, which are frequently associated with a risk of morbidity or mortality. [3][4][5] These risks are of particular concern for tumors located within the brainstem, such as DIPG, where biopsy is technically very challenging and is associated with a risk of significant complications. 3,4 The study of liquid biopsy has emerged as an alternative and/or complementary approach to tumor biopsy. Liquid biopsy analysis is comprised of the study of tumor derived material from any biological fluids including blood, cerebrospinal fluid (CSF), urine, and saliva. In this context, cell-free DNA (cfDNA), extracted from different biofluid sources, is used to assess tumor-specific alterations in a less invasive manner. The fraction of cfDNA derived from tumor cells is known as circulating tumor DNA (ctDNA). A benefit of liquid biopsy analysis is the ability to correlate the presence of driver mutations with tumor burden and response to therapy at multiple time-points, avoiding the risks, costs, and need for the expertise of surgical intervention. In this context, many pediatric gliomas are characterized by hotspot driver mutations (H3.1/H3.3_K27M, H3.3_G34R/V, BRAF_V600E, IDH1_R132H) 6 or by single fusion events (ETV6:NTRK3, KIAA1549:BRAF). 7,8 This makes them perfect candidates for the use of ctDNA to monitor treatment response enabling early detection of tumor progression over the course of the disease.
The presence of ctDNA from plasma has been demonstrated in a range of pediatric solid tumors. [9][10][11][12][13][14] Despite concerns regarding the utility of such approaches in brain tumors, 15 several studies have illustrated that ctDNA can be detected in CSF from a variety of pediatric and adult central nervous system (CNS) malignancies. [16][17][18][19][20] In particular, Wang and colleagues found molecular alterations in 74% of patients from ctDNA derived from CSF, obtaining an average of 417 ng of ctDNA in an average of 4.8 mL of CSF using amplicon next-generation sequencing approaches (NGS) methodology (SafeSeqS). 16 Similar results have been observed in terms of detectable mutations in ctDNA derived from CSF in patients with pHGG and DIPG by using nested polymerase chain reaction (PCR), droplet digital PCR (ddPCR), and panel sequencing. [21][22][23][24] These studies have also demonstrated that CSF-derived ctDNA levels increase during disease progression. 22,23 In adult glioma, use of the MSK-IMPACT capture-based NGS assay identified ctDNA in CSF from around half of patients, with ctDNA levels correlating with disease burden and poor outcome. 25 Less is known about the utility of plasma-derived DNA from brain tumors toward mutation detection and treatment response monitoring, and in particular in pHGG, DIPG, and other DMG. Pan and colleagues showed the detection of ctDNA by custom capture panel sequencing, derived from 3 mL of plasma in 3/8 pediatric patients with brainstem tumors; of those three, two had undetectable mutations in the plasma ctDNA compared to the ctDNA derived from the CSF. 24 Conversely, a study from Panditharatna and colleagues showed detectable levels of ctDNA derived from 1 mL of plasma in 16/20 patients with DMG at diagnosis using ddPCR. 23 Notably, a recent publication described the use of cell-free methylated DNA immunoprecipitation and high-throughput sequencing (cfMeDIP-seq) on ctDNA derived from plasma in a range of adult glioma specimens. 26 Given the importance of methylation-based profiling for subtyping pediatric diffuse glioma, this would be an important technique to assess in the childhood context.

Importance of the Study
Children with high-grade and diffuse midline glioma have an invariably fatal outcome, and with surgical resection impossible when occurring in the brainstem (diffuse intrinsic pontine glioma, DIPG), such non-invasive specimens have the potential to play a vital role in tumor diagnosis and disease monitoring. Here we show the utility of sensitive and specific ddPCR techniques to reliably detect circulating tumor DNA of all major subtypes of pHGG/DMG (including DIPG), from plasma, serum, cystic fluid, and CSF. We further screened very limited quantities of serial plasma samples collected as part of the HERBY clinical trial to show that cell-free DNA can be used to track progression in both hemispheric and midline tumors. These data provide a rationale for the incorporation of such liquid biopsy collection into future clinical trials for inclusion and molecular stratification, monitoring of treatment response, and for guiding novel therapeutic interventions at relapse.

Neuro-Oncology Advances
The implementation of ctDNA from plasma and CSF in routine clinical practice represents an important goal for the field. The inclusion criteria for an increasing number of clinical trials require molecular characterization to confirm biomarker positivity. For example, H3F3A_K27M and BRAF_ V600E need to be confirmed in tumors for patients to be eligible for current clinical trials of ONC201 (NCT03416530) or dabrafenib in combination with trametinib (NCT02684058), respectively. This is of particular importance for patients such as those with DIPG, where tissue biopsy remains an invasive procedure, not without complications. 3,4 In addition, the analysis of ctDNA can provide a unique opportunity to assess therapeutic response to a targeted agent, as well as to track tumor evolution in response to therapy, and to identify potential resistance mechanisms that may inform novel treatment options at relapse.
To this end, we sought to explore whether molecular alterations could be identified in liquid biopsy samples from pHGG, DIPG, and other DMG patients. ddPCR assays were validated and applied to quantify ctDNA levels derived from plasma, serum, and CSF. We also explored whether circulating DNA concentrations correlated with tumor burden and multimodal radiological indicators of response and tumor progression using samples collected within a clinical trial in nonbrainstem pHGG/DMG.

Cases
All patient samples were collected after signed consent to either the HERBY or BIOMEDE translational research programs, or local Institutional Research Board, under full Research Ethics Committee approval at each participating center. A total of 44 samples from different source of liquid biopsy sample sources (plasma n = 27, serum n = 6, CSF n = 10, and Cyst fluid n = 1), were collected from local studies (Royal Marsden Hospital n = 26) and collaborators (Ospedale Pediatrico Bambino Gesù n = 8 and BIOMEDE n = 10). Samples corresponding to 33 patients harbored mutations identified in the tissue tumor sample by nextgeneration sequencing (whole-exome sequencing and capture panel sequencing assays [27][28][29] ). In addition, 127 plasma aliquots from different time-points, were available from 41 HERBY patients, harboring driving mutations identified by whole-exome-sequencing of pre-treatment tissue tumor samples. 29

Liquid Biopsy Samples
Where possible, up to 10 mL of peripheral blood was collected into Cell-Free DNA Collection Tubes (Streck, La Vista). Samples were centrifuged twice for 10 min, first at 1600 g and at up to 16,000 g to remove cellular contents and/or debris. Samples were stored at −80°C until cfDNA extraction. Local protocols to isolate plasma and CSF were used for the remaining liquid biopsies cases, collected from different sources. cfDNA isolation from plasma and CSF supernatant was performed using the QIAamp circulating nucleic acid kit (Qiagen, 55114) following quantification using the Qubit fluorometer (ThermoFisher Scientific, dsDNA HS Assay kit, Q32854) and fragment analysis by 2200 and 4200 TapeStation (Agilent, Genomic DNA ScreenTape 5067-5366).

Droplet Digital PCR
Custom TaqMan-based quantitative PCR genotyping assays (Applied Biosystems, Thermo Scientific and IDT, Integrated DNA Technologies) were designed to specifically detect genetic abnormalities (mutations) (Supplementary Table S2). Commercially available assays were used to identify MYCN amplification (MYCN Hs00201049_cn, control region 4403316 or 4403326, Applied Biosystems, Thermo Scientific) and H3F3A_K27M (H3F3A_K27M dHsaCP2500510; H3F3A_WT dHsaCP2500511, Bio-Rad) as well as H3F3A_G34R (H3F3A_G34R dHsaIS2502308; H3F3A_WT dHsaIS2502309, Bio-Rad). The assay limit of detection (LoD) was assessed by performing serial dilutions of the mutant DNA in a constant concentration of wildtype DNA (1:10, 1:100, 1:1000 and 1:10,000) and run in duplicate using 5 ng of DNA. The LoD was calculated as the fractional abundance of the neat mutant sample divided by the lowest dilution with detectable mutant copies (at least two mutant droplets). 30,31 For each assay, three controls were run in duplicates including: one non-template control, one wild-type control (fragmented Promega DNA at 1 ng/µL), and one positive control harboring the alteration of interest.
The Bio-Rad QX200 ddPCR system was used, which allows the detection of rare DNA target copies with high sensitivity. DNA was randomly encapsulated into approximately 15,000 oil nanoliter-sized droplets, using the Automated Droplet Generator (BioRad, QX200 AutoDG), containing ddPCR Supermix for probes (no dUTP) (BioRad, 1863024), genotyping assay (specific per alteration), water, and the DNA of interest. The PCR reaction was performed in a thermocycler plates were then placed on the droplet reader where the droplets are streamed individually through a detector and signals from mutant positive (FAM), wild-type (VIC/HEX), double-positive (FAM and VIC/ HEX), and negative droplets (empty) are counted to provide absolute quantification of DNA in digital form. The mutant allele concentration (C MUT ) and wild-type allele concentration (C WT ) were calculated with Quantasoft Analysis Pro (BioRad), the mutant allele fraction (AF dPCR ) and the concentration of cfDNA in the CSF or plasma (CcfDNA ng/ mL) were calculated with the following calculations as previously described in 17,18,22 : the mass of 1 haploid human genome is 0.003 ng

Fusion Panel
A custom fusion panel consisting of 24 genes associated with fusions in pediatric brain tumors (ALK, BCOR, BEND2, BRAF, c11orf95, C19MC, CIC, ETV6, FGFR1-3, FOXR2, QKI, KIAA1549, MET, MN1, MYB, MYBL1, NTRK1-3, RAF1, RELA, TPM3, and YAP1) was designed with a library of probes to ensure adequate coverage of the specified regions (Roche Sequencing Solutions). 29 30 ng of cfDNA was used for library preparation using KAPA HyperPlus Kit (Kapa Biosystems) and SeqCap EZ adaptors (Roche) without performing the fragmentation step. DNA was end-repaired, A-tailed, and indexed adaptors ligated, amplified, multiplexed, and hybridized using 1 μg of the total pre-capture library DNA. After hybridization, capture libraries were amplified and sequencing was performed on a MiSeq (Illumina). Quality control (QC), variant annotation, deduplication, and metrics were generated for each sample. The raw list of candidates provided by Manta (https://github.com/Illumina/ manta) were filtered for more than 2 reads covering both genes, common false-positive base pairs (bp) positions/ fusions outside of the capture set at both ends, common breakpoint/false-positives within 10 bp, common false positive gene pairs, fusions within the same gene and homologous sequences greater than 10 bp.

Radiological Evaluation
Analysis of tumor burden from the HERBY cohort was carried out at different time-points based on imaging and clinical data. Following image review by up to three expert pediatric neuroradiologists on the HERBY Central Radiology Committee using the Response Assessment in Neuro-Oncology (RANO) criteria, 32 an independent pediatric oncologist reviewed supportive clinical data and corticosteroid dosage and provided the final status for that time point. [29,30]

Statistical Analysis
Statistical analysis was carried out using GraphPad Prism 8, using one-way ANOVA with multiple testing correction. An adjusted P-value of less than .05 was considered significant.

ddPCR Assay Validation for the Detection of ctDNA from Liquid Biopsies
Liquid biopsies from multiple biological sources (plasma, serum, cerebrospinal fluid [CSF], and cystic fluid) were collected from 32 pHGG and DIPG patients with known molecular alterations from the sequencing of their tumor tissue. These patients harboured somatic mutations in H3F3A (K27M and G34R), BRAF (V600E), ACVR1 (G328V), IDH1 (R132H and R132S), TP53 (C238Y and R282W), and PIK3CA (E542K and H1047R), and one had MYCN amplification. The first goal was to develop a robust detection method for these genetic alterations. To do this, customized and commercially available assays for ddPCR were validated for the identification of patient-specific molecular alterations. Each genotyping assay was tested by using a positive sample harboring the specific alteration of interest, and the variant allele frequency (VAF) was compared between ddPCR and NGS, with an observed correlation of r 2 = 0.9543 ( Figure 1A).
To assess the limit of detection (LoD) of point mutation detection assays, mutant DNA samples were serially diluted 10-fold in wild-type genomic DNA (1/10, 1/100, 1/1000, and 1/10,000). Genomic DNA from tissue was fragmented and a total DNA input of 5 ng was utilized to simulate the anticipated low amount of ctDNA. LoD was calculated as the VAF of the neat sample divided by the lowest dilution with detectable signal for mutant, with at least two droplets containing mutant DNA. Two different H3F3A_K27M assays were assessed, one commercially available from Bio-Rad and one reported by Stallard and colleagues. 22 Both assays performed well, obtaining a good droplet separation between FAM and VIC/HEX labels, with a similar LoD (Bio-Rad = 0.793% and custom = 0.791%) ( Figure 1B-E). In addition, no mutant droplets were observed in any of the wild-type template control DNA included per assay in each run. By using 5 ng of DNA, LoD ranged from 0.041% (PIK3CA_E542K) to 0.993% (TP53_C238Y), with a median of 0.203% (Supplementary Table S1) and (Supplementary Figure S1).
The MYCN amplification assay contained two probes, one within the MYCN gene and one in a control region at chromosome 5p15. 33. The amplification assay was tested by comparing the ratio of copies/µL of MYCN to the control gene (Supplementary Figure S2A). Two ctDNA-plasma positive samples from MYCN-neuroblastoma patients were used for the assay validation, with tissue samples for each patient used as a positive control and run in duplicate.
Finally, a custom pediatric brain tumor fusion panel 29,33 was used to detect a known ETV6:NTRK3 fusion in the CSF from a single infant glioma patient (OPBG_INF_035). 30 ng of cfDNA extracted from 4.5 mL of CSF was run on the capture panel, with 23 reads supporting the fusion detected (Supplementary Figure S3).

Exploring the Use of Liquid Biopsy in the HERBY Clinical Trial Cohort
To assess the utility of liquid biopsies for molecular diagnostics and to monitor disease progression, we studied genetic alterations in cfDNA derived from plasma from longitudinal samples from the well-annotated HERBY trial in non-brainstem pHGG (NCT01390948). 29,34-37 Blood samples were taken at up to five different time-points during the course of treatment, with plasma isolated locally and sent to our laboratory. cfDNA was extracted from 127 plasma samples from 41 patients, selected for tumors harboring alterations in H3F3A, IDH1, BRAF, or MYCN ( Figure 3A).
The mean volume of plasma from which cfDNA was extracted was 0.49 mL (SD = 0.35, excluding one sample from which 4 mL of plasma were used for extraction) ( Figure 3B). The mean yield of total DNA extracted from plasma was 2.52 ng (SD = 2.83, excluding the four cases with high levels of genomic DNA) ( Figure 3C). The mean of total DNA yield extracted per mL of plasma was 5.25 ng (SD = 5.21, excluding the four cases with high levels of genomic DNA) ( Figure 3D). The DNA samples were run undiluted and the mean of DNA ddPCR input was 1.76 ng (SD = 2.04). Disappointingly, none of the HERBY cfDNA samples tested for the known genetic alterations were

Neuro-Oncology Advances
positive (>two mutant droplets for point mutations and >4-fold for MYCN amplification). However, there were four cases where one positive droplet was found (BRAF_ V600E, n = 2; H3F3A_K27M, H3F3A_G34R, n = 1 each). All four patients received bevacizumab and had stable disease as their best radiological response. cfDNA concentration was compared between molecular subgroups. Although there was no significant difference between subgroups at baseline (P = .1026, one-way ANOVA), there was a trend of higher concentration of cfDNA in BRAF_V600E positive patients compared to H3F3A_ K27M and H3F3A_G34R (P = .0547 and P = .0661, respectively, one-way ANOVA, Dunnett's multiple comparisons test) ( Figure 3E). DNA integrity was measured by using TapeStation, showing four different types of DNA size distribution: 33 samples presented a detectable cfDNA peak (~170 bp) ( Figure 3F), five samples contained a high amount of genomic DNA contamination (>55 kb) ( Figure  3G), 12 samples showed detectable cfDNA and genomic DNA peaks ( Figure 3H), whilst in the remaining 75 samples no DNA was detectable ( Figure 3I).
Finally, although we were not able to reliably detect ctDNA in the HERBY plasma samples, we explored the correlation of cfDNA concentrations to disease burden and tumor progression. When assessing the changes in cfDNA concentrations over the course of the individual patient's disease (Supplementary Figure S4), anecdotal variations across longitudinal time-points were observed in four patients. Two DMGs, both H3F3A_K27M mutated, exhibited increased cfDNA concentrations at later timepoints, corresponding with a relatively short time to progression in these cases. The first (HERBY032) was a 12.8 year-old boy who underwent a near-total resection prior to treatment with bevacizumab and chemoradiotherapy. He displayed local recurrence at 5.5 months, though there was a marked increase in cfDNA concentration 3 months earlier.
He died at 16.4 months postrandomization ( Figure 4A). The second (HERBY096) was a 12.6 year-old boy, also on the bevacizumab arm, but who was eligible for biopsy only, and had local progression at 4.0 months. There was a substantial cfDNA increase in the subsequent plasma sample two months later, and he died of disease at 8.7 months ( Figure 4B).
Conversely, two hemispheric glioblastomas with BRAF_V600E mutations showed a reduction in cfDNA concentration from baseline and early time-points, corresponding to longer progression-free survival. HERBY063 was a 10.5-year-old boy who underwent three resections and survived for 28.5 months postrandomization to bevacizumab plus chemoradiotherapy. There was a marked decrease in cfDNA concentration at the earliest timepoints in the first two months, and prior to later local recurrence with slow growth at 8 months ( Figure  5A). Finally, HERBY078 was a 13.8-year-old girl treated with temozolomide and radiotherapy alone, and also displayed a substantial initial decrease in cfDNA. She progressed at 10 months locally and below the skull base, with evidence of parotid gland metastatic spread. Nonetheless, she survived on treatment for 27.4 months before succumbing to her disease ( Figure 5B). Shown are the mutations assessed by ddPCR, the biological source material, volume used for the ddPCR assay, variant allele frequency (VAF), number of mutant and wild-type droplets as well as concentration of mutant and wild-type cfDNA.

Discussion
This study describes the validation of a number of ddPCR assays for the detection of point mutations in cfDNA. These include key genes commonly altered in pHGG, DIPG, and other DMG, including H3F3A, IDH1, PIK3CA, BRAF, ACVR1, and TP53, as well as amplification of MYCN. By applying this methodology to cfDNA, tumor mutations were detectable in CSF, cystic fluid, plasma, and serum derived from pHGG and DIPG patients. In accordance with other studies, it was found that ctDNA was present at a higher percentage and with greater VAFs in cfDNA derived from CSF compared to plasma and/or serum specimens (~67% compared to 26% and 33% samples, respectively); these data support the use of CSF over plasma as source of tumor DNA for molecular profiling. 17 Circulating tumor DNA represents a small fraction of total cfDNA, and the low yields seen in pHGG and DIPG patients represent a major challenge for the detection of this potentially useful biomarker. It is thought that the low permeability of the brain-blood barrier might prevent ctDNA from spreading into the bloodstream. This is supported by the fact that higher ctDNA levels derived from plasma are observed in patients with diffuse midline glioma after radiation (72-100 hours), suggesting that radiotherapy might disrupt the BBB allowing ctDNA to be released into the bloodstream. 22,23 Another possible reason for lower levels of ctDNA isolated from plasma and or serum is the presence of background genomic DNA from non-malignant cells. In particular, it was observed

Neuro-Oncology Advances
that no ctDNA was detected in samples presenting highly fragmented cellular DNA, presumably derived from cells undergoing necrosis. Samples presenting higher levels of genomic DNA in our cohorts were mostly derived from external institutions where blood samples were not taken using collection tubes containing a preservative stabilizer of nucleated blood cells such as Streck or PAXgene blood ccfDNA tubes. The use of these tubes is highly recommended to prevent cell lysis, and when this is not possible samples taken in EDTA tubes should be processed within 2 h of blood withdrawn.
Longitudinal plasma samples from HERBY, the largest randomized clinical trial in non-brainstem pHGG, represented a unique cohort to test the utility of such approaches for disease monitoring. 29,[34][35][36][37] Unfortunately, at the time of study initiation in 2011, the protocol allowed for only small fluid volumes to be taken, as liquid biopsy approaches had not yet been considered, and it is unfortunate that no ctDNA could be detected from such limited amounts. This is an important consideration for future trials, with at least 4 mL of plasma required for liquid biopsy tests used in clinical practice such as Guardant360. 40 Despite this, cfDNA concentrations themselves, when detectable above baseline, correlated with early disease progression and poor outcome in two patients with K27M mutated DMG, and a better outcome for two patients with BRAF_V600E mutated hemispheric GBM patients.
As cellular DNA contamination can affect the sensitivity of ctDNA detection, some studies have applied in-silico and in vitro size selection to a achieve higher sensitivity evaluation of ctDNA. 41 However, this needs to be further verified as size selection after cfDNA extraction might contribute to potential loss of ctDNA material. Another strategy that Panditharatna and colleagues used in their study, which detected ctDNA in 80% of diffuse midline gliomas at diagnosis/upfront therapy, was a pre-amplification step of 9 cycles. 23 This could explain their high detection rate and should be further validated to assess the potential false positive rate introduced by pre-amplification. Newer strategies combining the use of unique molecular identifiers (UMIs), to facilitate the identification of single DNA molecules from PCR duplicates, with deep sequencing, are promising strategies to detect ctDNA. 42,43 In addition, this strategy sequences a list of genes that can be customized allowing the detection of multiple genes, which can be valuable to track emergence of resistance alterations. In this context, Cell3 Target (Nonacus, oncology) offers calling of mutations down to 0.1% of VAF from as little as 10   Resections are marked with an X. Below, axial T2-weighted MRI scans at different time-points of the patient disease, with white arrows highlighting a new parotid lesion, and the black arrow indicating the primary site recurrence. The shaded box represents the initial 6-week treatment of RT/TMZ. Subsequent to this, there were repeated cycles of TMZ every 28 days until the end-point.

Neuro-Oncology Advances
input by incorporating UMIs into targeted NGS customised gene panel.
In summary, we could identify tumor-specific DNA alterations more readily in CSF than plasma, demonstrating the feasibility of tracking tumor response, but also highlighting the importance of sufficient plasma volumes and additional techniques that could enhance yield in these samples. This is particularly critical to avoid the risks associated with repeated sampling of CSF for serial monitoring over time in children with this disease.

Supplementary Data
Supplementary data are available at Neuro-Oncology Advances online.

Funding
This work was supported by Christopher's Smile, the CRIS Cancer Foundation, Abbie's Army, and Cancer Research UK. M.V. is a Children with Cancer UK Fellow. The HERBY clinical trial was carried out as a collaboration between the European Innovative Therapies for Children with Cancer (ITCC) consortium, the SIOPE Brain Tumour Group, the Australian Children's Cancer Trials Group (ACCT), and the C-17 Council (Canada).